On Sat, Jan 31, 2015 at 12:30 PM, Felix Schumacher < [email protected]> wrote:
> Am 30.01.2015 um 13:31 schrieb Philippe Mouawad: > >> Hi, >> I intend to commit a BackendListener client implementation that computes >> APDEX at end of Load Test. >> It will take: >> >> - Acceptable Response Time (http://www.apdex.fr/) T >> > The link was in french (my french is not very good (really non existant > :)), but luckily wikipedia had an article about apdex, which I could read. > Sorry for the french link some references: http://www.apdex.org/overview.html - Compute F as 4xT but allow customization >> - List of samples taken into account >> >> >> This listener will compute it during load test and generate an HTML file >> at >> end of test containing it. >> >> This listener could be later enhanced to : >> >> - Generate equivalent of Aggregate Report >> - Possibly graphs based on chartjs or some other graphing JS Library >> > I understand, that apdex is a simple mean to have one metric. As such I > wonder why we would want to graph it. In fact my proposal was larger than just APDEX. What I am proposing is to implement a report generator that at end of Test would compute HTML page with: - APDEX - Some graphs like: - Plot of OK/KO (http://www.chartjs.org/docs/#doughnut-pie-chart) - Request Summary (what's currently in Aggregate Graph) http://www.chartjs.org/docs/#bar-chart - Response Time over time (http://www.chartjs.org/docs/#line-chart) I think all this should in fact be computed at end of load test instead of during it. The input of this listener would be the generated CSV file, and at end of Test it would read it an compute all this. > If we want to display a graph of apdex values for different tolerated > times, than we would have to store all values, which would make the > listener quite heavy on the memory side. > >> >> Thoughts ? >> > How will errors be counted? Sometimes I can cope with errors, as long as > they are reported fast and sometimes an error would result in a very > unhappy consumer. > Good catch, I think only successful samples must be taken into account. Error will count only toward total requests (unsatisfied). > > Regards > Felix > > -- Cordialement. Philippe Mouawad.
